Remove unrolling of last iteration of loop for x86 decode (#80)

* Remove unrolling of last iteration of loop for x86 decode

It is not clear why this was originally done. There is no comment
in the source code, but it does not seem to improve performance.

* Load keys for current loop iteration instead of next iteration

Author: andrewthad

src/streamvbyte_x64_decode.c

@@ -48,39 +48,9 @@ static inline const uint8_t *svb_decode_sse41_simple(uint32_t *out,
     int64_t Offset = -(int64_t)keybytes / 8 + 1;
 
     const uint64_t *keyPtr64 = (const uint64_t *)keyPtr - Offset;
-    uint64_t nextkeys;
-    memcpy(&nextkeys, keyPtr64 + Offset, sizeof(nextkeys));
-    for (; Offset != 0; ++Offset) {
-      uint64_t keys = nextkeys;
-      memcpy(&nextkeys, keyPtr64 + Offset + 1, sizeof(nextkeys));
-
-      Data = svb_decode_sse41((keys & 0xFF), &dataPtr);
-      svb_write_sse41(out, Data);
-      Data = svb_decode_sse41((keys & 0xFF00) >> 8, &dataPtr);
-      svb_write_sse41(out + 4, Data);
-
-      keys >>= 16;
-      Data = svb_decode_sse41((keys & 0xFF), &dataPtr);
-      svb_write_sse41(out + 8, Data);
-      Data = svb_decode_sse41((keys & 0xFF00) >> 8, &dataPtr);
-      svb_write_sse41(out + 12, Data);
-
-      keys >>= 16;
-      Data = svb_decode_sse41((keys & 0xFF), &dataPtr);
-      svb_write_sse41(out + 16, Data);
-      Data = svb_decode_sse41((keys & 0xFF00) >> 8, &dataPtr);
-      svb_write_sse41(out + 20, Data);
-
-      keys >>= 16;
-      Data = svb_decode_sse41((keys & 0xFF), &dataPtr);
-      svb_write_sse41(out + 24, Data);
-      Data = svb_decode_sse41((keys & 0xFF00) >> 8, &dataPtr);
-      svb_write_sse41(out + 28, Data);
-
-      out += 32;
-    }
-    {
-      uint64_t keys = nextkeys;
+    uint64_t keys;
+    for (; Offset != 1; ++Offset) {
+      memcpy(&keys, keyPtr64 + Offset, sizeof(keys));
 
       Data = svb_decode_sse41((keys & 0xFF), &dataPtr);
       svb_write_sse41(out, Data);